Jobs are bundles
The insight: a job is a bundle
In a 2003 paper that now underpins most serious automation analysis, David Autor, Frank Levy, and Richard Murnane argued that no job is a single thing. Every job is a bundle of tasks — some routine, some not, some cognitive, some manual. Technology does not replace jobs; it replaces tasks, and the bundle changes around the remaining human work.
Why the framing matters
Saying 'AI will take jobs' is a category error. What AI will do — what any technology does — is change the composition of tasks inside jobs. Some tasks get automated away; others get cheaper; new ones appear. A 'radiologist' in 2026 does fewer of the same tasks than a radiologist in 2006, but more new ones (supervising the AI, managing edge cases, communicating with patients).
Routine vs non-routine
Autor's framework distinguishes routine tasks (following a clear rule, cognitive or manual) from non-routine tasks (judgment-heavy, unpredictable). Historically, routine tasks were automated first. The big change since 2020 is that many non-routine cognitive tasks — drafting, summarising, designing — have also become machine-doable.
Implications for workers
If jobs are bundles, then two strategies make sense. First: learn which of your own tasks are most exposed, and invest in the parts of your bundle that aren't. Second: learn to use the automation tools themselves, so you keep more of the gains rather than less. Both strategies will recur throughout this course.
Write down five tasks that make up your own work — or someone else's you know well. Which could a 2026-era AI do today?